Every month, half a million new research papers are published – give or take a few thousand. The 25 to 30 million authors – give or take a couple hundred thousand – are researchers dedicated to their individual fields of study. That’s a lot of brilliant minds releasing a seemingly never-ending well of knowledge. Sadly, however, much of their work gets lost in academia or locked behind a paywall.
Referring to these untapped research papers as “the biggest waste of potential in modern times,” Geet Khosla, co-founder and CEO of Proem, is setting out to change that. With the launch of Proem, users can get answers to their questions based on actual scientific research that is being published on what Khosla has described as the largest repository of scientific research in the world.
To comb through the research, summarize the findings and help non-experts understand complex topics, Proem leverages the most powerful AI models from OpenAI, Anthropic, Meta and Google. And it’s all delivered on a platform that has the look and feel of a social media feed.
But what does that mean for manufacturers? For starters, it’s important to note that all that research isn’t just happening in traditional fields like medicine or physics. Researchers are publishing on economics, material science, aerospace engineering and beyond. And that equates to manufacturing insight to keep businesses competitive and nimble for upcoming technology breakthroughs or shifts in the industry.
To learn more about the new tool, Shop Floor Lasers got the chance to speak with Khosla. A snapshot of that conversation reveals a tool that puts ChatGPT on its heels.
Shop Floor Lasers: How did you source the research papers?
Khosla: Every piece of research that’s been written typically sits on a pre-print server that’s really well-hidden except for people like us that can find it. And that research can sit there for quite a while – academic journals just can’t handle the volume of papers being released. Currently, we have access to more than 240 million research papers on our platform.
Essentially, the only difference between a published paper and one that’s found on a pre-print server is peer review. The peer review process, however, isn’t about the experiment or the findings; it’s more about how the paper is written. The novel insight doesn’t change – the only thing that does change is the cost of the paper. Post peer-review, access can come with a price tag anywhere from $500 to a few thousand dollars.
So that was our aha moment. We recognized that we could access the abstracts of the latest research papers that are publicly available and then extract the core insights out of them. Most people have a specific need or a specific question and, therefore, only need the core insight from a paper. Many users on Proem are simply exploring a topic to discover the data sources or the credible sources of writing that are out there. Besides, you might not have the time, the energy or the need to read an entire 50-page paper.
And as far as the journals are concerned, we collaborate with them to let them know when something is trending. From there, they will often bump it up on the peer review ladder and try to get it reviewed faster.
SFL: How does Proem differ from ChatGPT?
Kosla: ChatGPT and the other AI model answer engines and chatbots base their answers on the model’s training. You can ask questions in interesting ways, but it will give you very different answers each time. It’s not trying to give you an answer based on a corpus of data; it’s trying to give you an answer based on your question.
ChatGPT is great as a draft writer and as a collaborator, but if you want to make sure that what you’re going to share is trustworthy, that the knowledge has come from a credible source, it’s unreliable. That’s not what these engines are made for.
Conversely, we take the best large language models out there and have built a way to connect every answer to a research paper. Every single word you read on our platform is based on a piece of research or multiple research papers. We’re not a generalist tool. We’re a tool that gives you knowledge at a general level, but from scientific research. That’s our purpose.
SFL: What is the user experience like? Can it be compared to a social media feed?
Khosla: Definitely. We generate a hyper personalized feed, so everyone’s feed will be different. What you might recognize as a social media post is actually coming from an individual paper where we take out a core insight and present it as a very simple, one- or two-sentence headline of sorts. You can explore and see what you’re interested in and then dive deeper into that.
To create a feed, we have what we call an assistant that compiles all of the questions you’ve been asking to give you answers based on all of the general research available.
Users can also tap into the research that institutions around the world are putting out. People don’t always recognize this, but companies like Nvidia put out a lot of research on a regular basis, so we’ve created a feed of their work. As an example, Nvidia recently published a paper about circuit configurations. Users can ask the assistant direct questions about that paper and then dive deeper into the topic.
Also, when users are interested in a specific topic or researcher and would like to read the entire research paper, then there’s a link to it on the pre-print server.
SFL: Why should someone in manufacturing check it out?
The more niche the industry, the more we like them. Take the manufacturing industry as an example where laser welding stainless steel might be of interest. I haven’t personally read about that topic, but there’s probably plenty of research out there on it. Perhaps a user could provide examples of papers that have been published on laser welding to let the assistant know what they’re looking for in order to search the hundreds of millions of papers to get the latest on that super-specific area.
This high-level research on specific technologies can be incredibly powerful if someone just invested in or is about to invest in new equipment or a new process. With Proem, there’s no need to read lengthy research papers; users just ask the questions they’re interested in. It can save manufacturers a lot of time while helping them predict the next innovations that can change their business for the better.
SFL: How else could a business use Proem?
Everything we talked about so far is just a single player experience where you’re discovering things on a personal level. But, we’re also focusing on an update called Private Spaces, which is also a feed, but one that’s shared across a team or community. Within these Private Spaces, everyone sees the same set of papers. Members of that Private Space can all ask questions about specific papers, but they can also see one another’s questions to comment and learn from each other. When you can see the set of questions that one of your colleagues asked, it can spark new, interesting ideas. It can be a great way to collaborate.
Going back to the example of laser welding stainless, each company can have their own completely private feed that their team can discuss around. Within Spaces, a Tier 1 or Tier 2 supplier to Boeing can stay on top of all of the research that Boeing has been putting out to prepare themselves for the next generation of engines, for example.
Collaborators can see the new reports that Boeing is putting out, which could instigate the consideration of a new technology or a new way of creating parts. And don’t forget that these feeds can also be tailored, so if Boeing published something that isn’t based in its manufacturing practices, it won’t show it to you because you’ve indicated that you only care about material science, for example. The implications are pretty interesting.
SFL: How much does it cost?
Khosla: We are free for anyone, and for organizations that want to create “Private Spaces,” we are running a pilot program and are open to collaborate with knowledge-intensive teams. The rest, however, will be public and free. The whole point is to get a large audience to start asking questions to open up the research.
Overall, we’re just trying to provide users with cutting-edge scientific research. To make that effective, we’re doing our best to give it to users in the most bite-sized manner possible. Our answers or summaries are never too long, and we try to keep the jargon out as much as we can to provide an experience that’s as easy as reading a text or going on X or Instagram.
